1. Ask safety question first
Is anyone in danger right now?
Separates true emergencies from urgent-but-safe issues and prompts immediate human escalation where required.
Is anyone in danger right now?
Where is the issue happening?
What is getting worse while we talk?
Is anyone in danger right now?
Where is the issue happening?
Triage urgent calls by phone. Ask safety first, collect location and risk details, tell callers to contact emergency services for immediate danger, and save an escalation summary.
{
"outcome": "Emergency call triage outcome",
"fields": [
{
"description": "caller name captured during the call.",
"name": "caller_name",
"required": true,
"type": "text"
},
{
"description": "phone number from caller id captured during the call.",
"name": "phone_number_from_caller_id",
"required": true,
"type": "text"
},
{
"description": "location captured during the call.",
"name": "location",
"required": true,
"type": "text"
},
{
"description": "risk type captured during the call.",
"name": "risk_type",
"required": true,
"type": "text"
},
{
"description": "danger present captured during the call.",
"name": "danger_present",
"required": true,
"type": "text"
},
{
"description": "urgency captured during the call.",
"name": "urgency",
"required": false,
"type": "text"
},
{
"description": "human escalation required captured during the call.",
"name": "human_escalation_required",
"required": false,
"type": "text"
}
]
}
AI: Hi, I am an AI call workflow for Emergency call triage. Is anyone in danger right now?
Caller: There is water coming through the ceiling.
AI: Where is the issue happening?
Caller: It is near the light fixture, so I need someone urgently.
Make the first screen and opening line match what the caller will actually get. For Emergency call triage, the promise should be narrow enough that a caller understands the purpose before sharing details or scanning a QR code. Avoid broad claims like "we can help with anything"; a specific promise produces cleaner calls and clearer follow-up.
Decide which fields are required before the call can be considered complete. A practical first version should capture caller name, phone number from caller id, location, then send a summary that the workflow owner can act on without replaying the call. If a field is not used for routing, qualification, scheduling, or review, remove it from the first launch.
Write down the cases that should not be automated. Use human review for danger to life or property, active flooding, fire, injury, lockout, gas smell, caller cannot stay safe so the workflow stays useful without pretending to handle every edge case. Review the first real calls before connecting higher-risk actions or expanding the workflow.
Use it when the call is repeatable, the team already knows the information they need, and the caller benefits from speaking instead of filling out a form. It works best for property maintenance, field service, after-hours support.
The first question should identify the caller goal and gather enough context to continue naturally. For this template, start with: "Is anyone in danger right now?". Keep follow-ups short so the caller does not feel like they are reading a form over the phone.
Save a structured result with caller name, phone number from caller id, location, plus a summary, completion status, and handoff reason when needed. The owner should be able to act on the result without interpreting raw transcript text.
Human handoff should trigger when the caller needs judgment, asks for a person, gives conflicting answers, or matches one of the workflow-specific rules: danger to life or property, active flooding, fire, injury, lockout, gas smell, caller cannot stay safe.
The template includes prompt, questions, output fields, sample transcript, handoff rules, and a live call entry point.